How to Convert an Excel File (XLSX) to Parquet

Why there is no direct XLSX to Parquet button

An XLSX file is a zipped bundle of worksheets, styles, formulas and merged cells. Parquet is a columnar table with one fixed schema. The conversion is not a format swap so much as a decision about which rectangle of cells is actually the table — which is why every reliable route goes through a flat CSV first.

The no-install route

1. In Excel, select the sheet you want and choose File → Save As → CSV UTF-8 (Comma delimited). Pick UTF-8 explicitly; the plain "CSV" option writes a legacy code page and mangles non-ASCII text.
2. Drop the CSV into the CSV to Parquet converter. Types are inferred, the output is Zstd-compressed, and everything runs locally in your browser.
3. Verify the result in the Parquet Viewer — check the column types before anything downstream depends on them.

Before step 1, make the sheet look like a table: one header row, no merged cells, no blank rows above the header, no totals row at the bottom, and no notes in a stray column to the right. Anything else becomes a phantom column or a row of nulls in the Parquet output.

Fixing the types Excel broke

Spreadsheets are untyped by habit, so the CSV usually needs a pass before it becomes a well-typed Parquet file. Drop the CSV into the SQL Workbench and shape it there:

SELECT
  CAST(order_id AS VARCHAR)                   AS order_id,
  strptime(order_date, '%d/%m/%Y')            AS order_date,
  CAST(replace(amount, ',', '') AS DOUBLE)    AS amount,
  nullif(trim(notes), '')                     AS notes
FROM 'orders.csv';

Three things this handles, all of them common:

• Locale formatting. A German or French workbook writes 1.234,56. Strip the thousands separator and normalize the decimal mark before casting, or the value silently lands as text.
• Ambiguous dates. 03/04/2026 is March 4 or April 3 depending on who saved the file. strptime with an explicit format removes the guesswork; never let inference decide this one.
• Codes that look numeric. Postcodes, SKUs and account numbers must stay VARCHAR, or 00123 becomes 123 and joins start failing.

Download the corrected result as CSV, run it through the converter, and the Parquet file carries real types.

If you do have a local toolchain

DuckDB (1.2 and newer) reads XLSX directly through its excel extension, so the CSV hop disappears:

INSTALL excel; LOAD excel;
COPY (SELECT * FROM read_xlsx('orders.xlsx', sheet = 'Orders'))
TO 'orders.parquet' (FORMAT PARQUET, COMPRESSION ZSTD);

The pandas equivalent needs openpyxl installed alongside it:

import pandas as pd

df = pd.read_excel("orders.xlsx", sheet_name="Orders", dtype={"order_id": "string"})
df.to_parquet("orders.parquet", compression="zstd")

Pass dtype= for every identifier column. Without it, pandas infers int64 for code columns and the leading zeros are gone before you can object.

Size limits, in both directions

A worksheet holds at most 1,048,576 rows, so any Excel export is small by Parquet standards and conversion is quick. The interesting limit is the return trip: a Parquet file with tens of millions of rows will not fit back into a sheet. When that is the goal, filter or aggregate first — the open Parquet in Excel guide covers that direction.